The inspectable layer behind Human Data OS and App Data OS, with source links listed for review. After a competitor lost four terabytes, you don't take a vendor's word: you check the tool, release evidence, license proof, and runtime data movement before you trust a privacy claim.
The Studios and toolkit list source links. Desktop, CLI, package, and hosted-preview availability require current external proof before they are marketed.
Rubric Studio and Robotics Studio sit behind Human Data OS. Agent Studio and Aura3D sit behind App Data OS. The Trust Toolkit sits behind both. Each is the primitive the professional product is built on.
Local runtime and package claims are proof-gated. AuraOne hosted begins when the problem becomes shared authorship, approval queues, or team governance.
Each one runs on disk and leaves a portable file — a folder a reviewer can diff, a regression suite your CI can run, a release card an examiner can read. These are the source-linked primitives Human Data OS and App Data OS are built on, the same parts running locally and inspectable line by line. When the EU AI Act's training-data provenance rules land in August 2026, you check the provenance yourself before you rely on it. Open the source first. Then install.
Connect, inspect, record, and replay MCP and A2A runs. Ingest OTEL spans. Export a regression suite your CI can run — so the same mistake is caught again.
Apache-2.0. Source link listed; package, release, checksum, desktop trust, and hosted preview proof are required before binary availability is marketed.
Local, file-based, git-friendly authoring for criterion-level evals. Author, test, calibrate, and export — every rubric a plain file your team can review.
Apache-2.0. Source link listed; package, release, checksum, desktop trust, and hosted preview proof are required before binary availability is marketed.
A local-first workbench concept for teleop and VLA episodes: scrub synchronized streams, review failures and recoveries, probe policies, and stage a reviewed release card.
Apache-2.0. Source link listed; package, release, checksum, desktop trust, and hosted preview proof are required before binary availability is marketed.
Eval manifests, regression banks, dataset and embodiment cards, contamination audits — packaged as GitHub Actions, CLIs, and SDKs. The provenance layer behind both Human Data OS and App Data OS, runnable in your own CI ahead of the EU AI Act's August 2026 provenance rules.
SSR-safe Liquid Glass surfaces and accessible UI primitives for React and Next.js — components, recipes, and a playground. The interface layer the Open Studios are built on.
Prompt-to-code scene kits, typed GLB and glTF assets, WebGL and WebGPU rendering, diagnostics, and static deploy checks. Browser-native 3D an agent can edit and a team can review — and your code stays editable source.
The 3D primitive behind App Data OS. Registry package, license, and static-host proof require current external evidence before availability is marketed.
Two SDK layers you keep. AuraGlass for the interface. Aura3D for browser-native scenes, typed assets, and deployable 3D apps. Editable source, your code. Install one command and start.
The AI-native 3D SDK for the web. Prompt-to-code scenes, typed GLB and glTF assets, scene kits, diagnostics, and static deploy checks. Registry and license proof required before package availability is marketed.
Scaffold a product viewer, a cinematic scene, or a browser-native 3D app in one command. Real GLB and glTF assets, and code that stays yours to edit.
SSR-safe Liquid Glass surfaces, accessible primitives, recipes, and a playground — the interface layer the Open Studios are built on. Registry and license proof required before package availability is marketed.
The libraries underneath every Open IDE. Each runs from a notebook, a CLI, or a GitHub Action — installed in one line, read in full before you trust it.
Risk taxonomy and lint pass for MCP server manifests.
Offline contract tests for A2A agent cards and task lifecycles.
Deterministic replay harness the agent cannot game.
Portable Markdown and JSON card for one agent run.
Turn OTEL and Phoenix GenAI spans into eval regression cases.
Read the source. Run it locally. Keep the weights. Open is the on-ramp to the layer you run, not the endpoint that locks you in. Bring AuraOne in when the problem becomes shared authorship, approval queues, or governance across a team.
